aboutsummaryrefslogtreecommitdiff
path: root/vendor/gmp-6.3.0/mpn/generic/divrem_1.c
blob: c13aa7915f347559d803a4420f5b29b724f61822 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
/* mpn_divrem_1 -- mpn by limb division.

Copyright 1991, 1993, 1994, 1996, 1998-2000, 2002, 2003 Free Software
Foundation, Inc.

This file is part of the GNU MP Library.

The GNU MP Library is free software; you can redistribute it and/or modify
it under the terms of either:

  * the GNU Lesser General Public License as published by the Free
    Software Foundation; either version 3 of the License, or (at your
    option) any later version.

or

  * the GNU General Public License as published by the Free Software
    Foundation; either version 2 of the License, or (at your option) any
    later version.

or both in parallel, as here.

The GNU MP Library is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
for more details.

You should have received copies of the GNU General Public License and the
GNU Lesser General Public License along with the GNU MP Library.  If not,
see https://www.gnu.org/licenses/.  */

#include "gmp-impl.h"
#include "longlong.h"


/* The size where udiv_qrnnd_preinv should be used rather than udiv_qrnnd,
   meaning the quotient size where that should happen, the quotient size
   being how many udiv divisions will be done.

   The default is to use preinv always, CPUs where this doesn't suit have
   tuned thresholds.  Note in particular that preinv should certainly be
   used if that's the only division available (USE_PREINV_ALWAYS).  */

#ifndef DIVREM_1_NORM_THRESHOLD
#define DIVREM_1_NORM_THRESHOLD  0
#endif
#ifndef DIVREM_1_UNNORM_THRESHOLD
#define DIVREM_1_UNNORM_THRESHOLD  0
#endif



/* If the cpu only has multiply-by-inverse division (eg. alpha), then NORM
   and UNNORM thresholds are 0 and only the inversion code is included.

   If multiply-by-inverse is never viable, then NORM and UNNORM thresholds
   will be MP_SIZE_T_MAX and only the plain division code is included.

   Otherwise mul-by-inverse is better than plain division above some
   threshold, and best results are obtained by having code for both present.

   The main reason for separating the norm and unnorm cases is that not all
   CPUs give zero for "n0 >> GMP_LIMB_BITS" which would arise in the unnorm
   code used on an already normalized divisor.

   If UDIV_NEEDS_NORMALIZATION is false then plain division uses the same
   non-shifting code for both the norm and unnorm cases, though with
   different criteria for skipping a division, and with different thresholds
   of course.  And in fact if inversion is never viable, then that simple
   non-shifting division would be all that's left.

   The NORM and UNNORM thresholds might not differ much, but if there's
   going to be separate code for norm and unnorm then it makes sense to have
   separate thresholds.  One thing that's possible is that the
   mul-by-inverse might be better only for normalized divisors, due to that
   case not needing variable bit shifts.

   Notice that the thresholds are tested after the decision to possibly skip
   one divide step, so they're based on the actual number of divisions done.

   For the unnorm case, it would be possible to call mpn_lshift to adjust
   the dividend all in one go (into the quotient space say), rather than
   limb-by-limb in the loop.  This might help if mpn_lshift is a lot faster
   than what the compiler can generate for EXTRACT.  But this is left to CPU
   specific implementations to consider, especially since EXTRACT isn't on
   the dependent chain.  */

mp_limb_t
mpn_divrem_1 (mp_ptr qp, mp_size_t qxn,
	      mp_srcptr up, mp_size_t un, mp_limb_t d)
{
  mp_size_t  n;
  mp_size_t  i;
  mp_limb_t  n1, n0;
  mp_limb_t  r = 0;

  ASSERT (qxn >= 0);
  ASSERT (un >= 0);
  ASSERT (d != 0);
  /* FIXME: What's the correct overlap rule when qxn!=0? */
  ASSERT (MPN_SAME_OR_SEPARATE_P (qp+qxn, up, un));

  n = un + qxn;
  if (n == 0)
    return 0;

  d <<= GMP_NAIL_BITS;

  qp += (n - 1);   /* Make qp point at most significant quotient limb */

  if ((d & GMP_LIMB_HIGHBIT) != 0)
    {
      if (un != 0)
	{
	  /* High quotient limb is 0 or 1, skip a divide step. */
	  mp_limb_t q;
	  r = up[un - 1] << GMP_NAIL_BITS;
	  q = (r >= d);
	  *qp-- = q;
	  r -= (d & -q);
	  r >>= GMP_NAIL_BITS;
	  n--;
	  un--;
	}

      if (BELOW_THRESHOLD (n, DIVREM_1_NORM_THRESHOLD))
	{
	plain:
	  for (i = un - 1; i >= 0; i--)
	    {
	      n0 = up[i] << GMP_NAIL_BITS;
	      udiv_qrnnd (*qp, r, r, n0, d);
	      r >>= GMP_NAIL_BITS;
	      qp--;
	    }
	  for (i = qxn - 1; i >= 0; i--)
	    {
	      udiv_qrnnd (*qp, r, r, CNST_LIMB(0), d);
	      r >>= GMP_NAIL_BITS;
	      qp--;
	    }
	  return r;
	}
      else
	{
	  /* Multiply-by-inverse, divisor already normalized. */
	  mp_limb_t dinv;
	  invert_limb (dinv, d);

	  for (i = un - 1; i >= 0; i--)
	    {
	      n0 = up[i] << GMP_NAIL_BITS;
	      udiv_qrnnd_preinv (*qp, r, r, n0, d, dinv);
	      r >>= GMP_NAIL_BITS;
	      qp--;
	    }
	  for (i = qxn - 1; i >= 0; i--)
	    {
	      udiv_qrnnd_preinv (*qp, r, r, CNST_LIMB(0), d, dinv);
	      r >>= GMP_NAIL_BITS;
	      qp--;
	    }
	  return r;
	}
    }
  else
    {
      /* Most significant bit of divisor == 0.  */
      int cnt;

      /* Skip a division if high < divisor (high quotient 0).  Testing here
	 before normalizing will still skip as often as possible.  */
      if (un != 0)
	{
	  n1 = up[un - 1] << GMP_NAIL_BITS;
	  if (n1 < d)
	    {
	      r = n1 >> GMP_NAIL_BITS;
	      *qp-- = 0;
	      n--;
	      if (n == 0)
		return r;
	      un--;
	    }
	}

      if (! UDIV_NEEDS_NORMALIZATION
	  && BELOW_THRESHOLD (n, DIVREM_1_UNNORM_THRESHOLD))
	goto plain;

      count_leading_zeros (cnt, d);
      d <<= cnt;
      r <<= cnt;

      if (UDIV_NEEDS_NORMALIZATION
	  && BELOW_THRESHOLD (n, DIVREM_1_UNNORM_THRESHOLD))
	{
	  mp_limb_t nshift;
	  if (un != 0)
	    {
	      n1 = up[un - 1] << GMP_NAIL_BITS;
	      r |= (n1 >> (GMP_LIMB_BITS - cnt));
	      for (i = un - 2; i >= 0; i--)
		{
		  n0 = up[i] << GMP_NAIL_BITS;
		  nshift = (n1 << cnt) | (n0 >> (GMP_NUMB_BITS - cnt));
		  udiv_qrnnd (*qp, r, r, nshift, d);
		  r >>= GMP_NAIL_BITS;
		  qp--;
		  n1 = n0;
		}
	      udiv_qrnnd (*qp, r, r, n1 << cnt, d);
	      r >>= GMP_NAIL_BITS;
	      qp--;
	    }
	  for (i = qxn - 1; i >= 0; i--)
	    {
	      udiv_qrnnd (*qp, r, r, CNST_LIMB(0), d);
	      r >>= GMP_NAIL_BITS;
	      qp--;
	    }
	  return r >> cnt;
	}
      else
	{
	  mp_limb_t  dinv, nshift;
	  invert_limb (dinv, d);
	  if (un != 0)
	    {
	      n1 = up[un - 1] << GMP_NAIL_BITS;
	      r |= (n1 >> (GMP_LIMB_BITS - cnt));
	      for (i = un - 2; i >= 0; i--)
		{
		  n0 = up[i] << GMP_NAIL_BITS;
		  nshift = (n1 << cnt) | (n0 >> (GMP_NUMB_BITS - cnt));
		  udiv_qrnnd_preinv (*qp, r, r, nshift, d, dinv);
		  r >>= GMP_NAIL_BITS;
		  qp--;
		  n1 = n0;
		}
	      udiv_qrnnd_preinv (*qp, r, r, n1 << cnt, d, dinv);
	      r >>= GMP_NAIL_BITS;
	      qp--;
	    }
	  for (i = qxn - 1; i >= 0; i--)
	    {
	      udiv_qrnnd_preinv (*qp, r, r, CNST_LIMB(0), d, dinv);
	      r >>= GMP_NAIL_BITS;
	      qp--;
	    }
	  return r >> cnt;
	}
    }
}